AI trained in Kazakh language and culture to help preserve historical heritage
Artificial intelligence is being trained to better understand the Kazakh language, culture and traditions, with a new 500-question benchmark designed to assess AI models’ knowledge of Kazakhstan’s history and cultural heritage, Qazinform News Agency reports.
Anar Kurmanzhankyzy, head of a joint project between the International Society “Qazaq Tili” and OpenAI, presented the initiative.
The AI Evaluation Benchmark Suite assesses large language models in areas including comprehension, grammar, naturalness of Kazakh speech, proverbs and idiomatic expressions, academic translation, literary translation from Kazakh into English, translation of children’s literature, safety and ethnographic knowledge. It was developed in Kazakh from the outset, taking into account the linguistic and cultural characteristics of the Kazakh language.
A separate ethnographic benchmark comprises 500 questions covering the history, traditions and culture of the Kazakh people. The GPT-5.6 Terra model achieved an accuracy rate of 35.92% in the test.
The benchmark is planned to be made available to the academic community and could be used to evaluate other language models.
The project also aims to help preserve Kazakhstan’s historical heritage. Researchers are developing AI tools based on optical character recognition (OCR) technology to digitise Kazakh-language archival materials and textbooks. The technology can recognise Kazakh-language text in newspapers, books, and PDF documents, identify page structures, and accurately extract text, including from multi-column pages.
Developers are also collecting Kazakh-language text and audio data for AI training. The audio dataset currently comprises 12,000 hours of recordings, all cleaned of background noise, music and other extraneous sounds. Transcription models adapted to Kazakh have brought speech-to-text accuracy to 98%, according to Anar Kurmanzhankyzy.
Earlier, it was reported that specialists had compiled a 14-billion-token database to improve ChatGPT’s responses in Kazakh. It contains materials from different periods of the language’s development, covering history, culture, education and science, as well as the linguistic heritage of the Kazakh diaspora abroad.
Qazinform News Agency previously reported that Kazakh and Azerbaijani are currently the fastest-growing languages in ChatGPT, according to Sarah Friar, chief financial officer of OpenAI.
